Professional Documents
Culture Documents
SUMMARY: Infrastructure systems in the United States are aging and considerable investment is needed to renew
and replace a significant proportion of the existing systems. Piping systems, which are used in many infrastructure
systems such as the distribution networks for utilities – water, sewage, gas, oil, etc., are very important in this
regard. Real time scheduling is an important and necessary task in the planning and execution of construction
projects. This is of particular importance in the installation of pipe systems, for which it is time consuming to plan
and coordinate between team members the detailed requirements and information for the generation of practical
installation schedules. During the installation stage, there can be delays or interference that could lead to the
failure of the initial schedule plan. Current approaches are time-consuming, not automated and do not provide
real-time schedules. Thus, the process is still fragmented and essentially manual, with inefficient information flow.
To effectively improve the installation schedule, current knowledge of the installation site situation is important,
with this knowledge being used to generate realistic schedules. Artificial intelligence (AI) maximizes the value of
data by learning from previous cases and facilitates decision-making by making the process smarter and
automatic. This paper proposes a new AI framework with machine learning (ML) and heuristic optimization
techniques for automating practical pipe system installation schedule generation and optimization. A BIM model
is used as reference to provide pipe system component information. A hybrid knowledge-based system is developed
to integrate data-driven knowledge base and site-driven knowledge base on pipe system installation. K-Nearest
Neighbor (KNN) and Graph Neural Network (GNN) ML techniques are adapted to map extracted components
with the installation activities and their requirements for installation based on knowledge obtained from industry
experts and piping codes. In addition, a heuristic algorithm is adopted to optimize the installation schedule.
Finally, an optimal installation schedule that minimizes overlapping activities, time and cost is suggested.
KEYWORDS: Artificial Intelligence, Machine Learning, Automation, Pipe Systems, Schedule, BIM
REFERENCE: Jyoti Singh, Chimay J. Anumba (2022). Real-time pipe system installation schedule generation
and optimization using artificial intelligence and heuristic techniques. Journal of Information Technology in
Construction (ITcon), Vol. 27, pg. 173-190, DOI: 10.36680/j.itcon.2022.009
COPYRIGHT: © 2022 The author(s). This is an open access article distributed under the terms of the Creative
Commons Attribution 4.0 International (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.
2. RELATED WORK
Researchers and construction-related project participants have developed technologies related to Artificial
Intelligence (AI) field to decrease the dependence level of expert in construction planning and schedule control
(Liu et al. 2018). Various studies have focused on planning and optimization of practical schedules in diverse
application areas. Pan et al. (2021) established an optimal real-time sequencing strategy based on simulation
optimization approach with unbiased gradient estimators for appointment scheduling of patient. Ho and Yu (2021)
applied KNN regression to ascertain optimal scheduling strategies for switching chillers and temperature settings
of a chiller system in lowering its carbon emission. Kalathas and Papoutsidakis (2021) uses stored-inactive data
from a company and uses data mining and applied machine learning techniques to create strategic decision support
to generate maintenance schedules. Seccai et al. (2021) proposed a new efficient framework for solving the optimal
TV promo scheduling problem by adopting machine learning (ML) models. Bandi and Gupta (2021) developed a
framework to solve staffing and scheduling problems in operating rooms using historical case data. Kong et al.
(2021) used greedy randomized adaptive search procedure for slot planning and truck scheduling. Zhao et al.
(2020) optimized the construction duration and schedule for robustness based on the hybrid grey Wolf optimizer
with a sine cosine algorithm. Hosseini et al. (2021) adopted pedestrian simulation model and GA for staged-
evacuation schedule optimization. Amer (2020) adopted an active learning-based annotation workflow and tool
where H(k) ∈ R n×d are the messages computed after k steps of the GNN; M is the message propagation function,
which depends on the adjacency matrix; H(k−1) message generated from the previous message-passing step; W(k) ∈
R d×d is a trainable weight matrix.
3.3. Spatial Constraint Analysis
Installation sequencing order of pipe systems based on spatial constraints is one of the major issues during the
installation of pipe systems. Arbitrary installation of pipe systems and components with inappropriate sequence
planning leads to delay, rework, higher installation costs, and congested sites.. Proper sequencing and adequate
space for installation are significant to efficiently install a pipe system. Spatial constraint analysis is performed on
pipe systems sharing common space by comparing the location of respective pipe systems and components. Spatial
constraints occur when the space between nearby pipe systems is less than the minimum distance (D Buffer) required
for straightforward installation of both pipe systems simultaneously without any spatial clash or uneasy conflict
during installation. To know whether components of pipe systems are sharing common space, this study
formulated local boundary approach and determined if there is any overlap between local boundary of respective
pipe system as shown in Fig. 4. The minimum and maximum 3D coordinates of a single pipe system is extracted,
and a buffer value is added to them and local boundary box for each pipe system is drawn. After drawing local
boundary box for pipe system, local boundaries are checked for any space clash and are then marked as overlapped
boundary and pipe systems are determined as constraint pipe systems and are further evaluated for spatial
constraint analysis.
where, T = total time required to install all the pipe systems, hours; pc = piping crew in-charge; N = total number
of piping crew available for installation; 𝑛𝑝𝑐 = total number of pipes and pipe fittings installed by piping crew
pc; t = time taken by plumbing crew to install a single pipe, hours; 𝑡𝑖𝑑𝑙𝑒 = time taken by plumbing crew in waiting
for installation, 𝑡𝑑𝑒𝑙𝑎𝑦 = time taken by plumbing crew due to delay
(𝑇𝑜𝑙𝑑 −𝑇𝑛𝑒𝑤 )
𝐾𝑏 𝑇𝑒𝑚𝑝
(2)
P=𝑒
where, P = probability function of the acceptance criteria, T old = initial time, Tnew = new neighboring time, Kb =
Boltzmann constant, Temp = current temperature
4. ILLUSTRATIVE EXAMPLE
To illustrate the proposed AI-based framework for realistic schedule optimization, a hypothetical piping project,
which is based on real project but modified for ease of modeling and to enable consideration of a broad range of
pipe system installation problems (Singh 2020) with 13 pipe systems is used (see Fig. 6, which shows the BIM
model depicting a typical plant room). AI techniques help in optimizing practical schedules by utilizing knowledge
relevant to the installation process of pipe systems based on historical data and real-time knowledge of current site
condition. Dynamo for Revit (Autodesk 2021) was used to automatically capture all the necessary geometric and
semantic information of the pipe systems and components from the BIM piping model as mentioned in Section 3,
with the information stored in an MS Excel spreadsheet for further utilization. Table 1 shows geometric
information about 3D space and dimension of information of structural and architectural components. The start
and end points of all the pipe systems in 3D space are included in Table 2.
S. No. Object (X, Y, Z) Minimum Coordinate value (mm) (X, Y, Z) Maximum Coordinate value (mm)
After all the necessary details are extracted from the BIM model, the hybrid knowledge-based system was tested
on deciding the suitable installation information about activity packages and resource productivity rates for the
illustrated BIM model pipe systems. The list of attributes for activity packages selection with respective weighting
values as assumed for this example are project type (7%), pipe system type (13%), weight of pipe material (50%),
diameter of component (30%). The list of attributes for piping crew productivity rate selection with weighting
values are project type (14%), pipe system material (15%), equipment type (33%), weight of pipe material (38%).
The result for favorable installation activity packages and piping crew productivity rates with respective
recommendation error as per the adapted KNN and GNN techniques are shown in Table 4 and Table 5 (Appendix).
Recommendation error refers to the error associated with the recommendation of information (activity packages
and productivity rates) depending on the list of attributes assigned for a particular project as per industry experts’
knowledge. For this example, hybrid knowledge base consists of a total of seven activity packages AP 1 to AP7,
which account for a total of 14 combination lists and six productivity rates PR 1 to PR6, with 23 combination lists
based on the respective list of attributes are evaluated. The value of different productivity rates is PR 1 = 0.32, PR2
= 0.38, PR3 = 0.42, PR4 = 0.45, PR5 = 0.3, PR6 = 0.44. The different activity packages and productivity rates are
assumed to be only associated with pipes in pipe system while all pipe fittings are considered to have constant
activity packages as APF and productivity packages as PRF with value as 0.3 h. The predecessor-successor
relationship between pipe system as per spatial constraint analysis is shown in Fig. 7(a). Fig. 7(b) and Fig. 7(c)
show the waiting time required to install constraint pipe systems simultaneously following spatial constraint
analysis rule using KNN and GNN techniques, respectively. The installation of successor pipe system must be
done after predecessor pipe system completed installation of clashed component in pipe system. The waiting time
refers to the time taken to install clashed component in predecessor pipe system for clash-free installation of the
constraint pipe system.
The schedule for all 13 pipe systems as shown in the BIM model (Fig. 6) is calculated using SA as per the
recommended activity packages and productivity rates of respective pipe system components as stated in Table 4
and Table 5 (Appendix) and spatial constraint analysis predecessor-successor relationship. The initial and final
temperature for temperature cycle is assumed to be 0.1 and 5, respectively. The minimum schedule time using
KNN to extract information from the hybrid knowledge-based system as generated by SA assuming there is no
delay is 34.02 h when only one crew is available, 17.46 h when two crews are available and 11.74 h when three
crews are available. The minimum schedule time using GNN to extract information form hybrid knowledge based
system as generated by SA assuming there is no delay is 33.79 h when only one crew is available, 16.82 h when
FIG. 7. Spatial constraint analysis (a) Precedence logic between pipe systems 1 to 13 (b) Waiting time (h) required
to install constraint pipe system simultaneously based on KNN (c) Waiting time (h) required to install constraint
pipe system simultaneously based on GNN
After comparing the schedule results from both KNN and GNN ML techniques, it was found that GNN gives
shorter schedule times than KNN. GNN performs better than KNN and can have large discriminating power among
options and produces high accuracy, if the GNN’s aggregation scheme is highly expressive and has higher number
of deciding attributes. The framework was also tested for any delays occurred on site. The delays occurring during
the actual installation of pipe system on site tends to make original generated schedule as redundant. Therefore,
new optimized installation schedule should be generated based on expected time required to fix delay and continue
with the installation process of pipe system restricted due to delay. The expected time needed to overcome a
particular delay can be considered as per developed hybrid knowledge data base. For illustration purposes, it was
assumed that during the installation of pipe 3 of pipe system 1 (PS1), a hold up occurred due to unavailability of
a component. The delay of 3 h is expected to occur while the installation is resumed. The delay will be affected to
the PS1 and all other pipe systems which are successor to uninstalled PS1 components (pipes and fittings) in spatial
constraint analysis predecessor-successor relationship. The next practical schedule with minimum time as
generated by SA assuming the 3 h delay occurred due unavailability of component of PS1 is 36.79 h when only
one crew is available, 18.5 h when two crews are available and 13.42 h when three crews are available. Table 3
shows the result of the proposed AI-based framework for realistic schedule optimization. The convergence time
curve of the proposed framework is shown in Fig. 8.
1 PS3, PS9, PS6, PS10, PS13, PS11, PS7, PS5, PS1, PS12, PS9, PS6, PS7, PS3, PS13, PS10, PS11, PS5, PS1,
PS4, PS2, PS8 PS12, PS2, PS8, PS4
Optimal time hours = 33.79 Optimal time hours = 36.79
Piping crew 1 – PS(t) Piping crew 2 – PS(t) Piping crew 1 – PS(t) Piping crew 2 – PS(t)
Compared with traditional approaches, the proposed framework notably reduces the time for the generation of
realistic schedule for installation of pipe systems on piping projects. An experienced engineer needs at least 150
minutes to generate the one schedule of 13 pipe system (Singh 2020), from extracting information, finding precise
spatial sequence, and generating the installation schedule but without considering any site delay. However, the
proposed automated framework took about 3 to 4 minutes, which represents 97% saving in time to generate
realistic schedule. As the built environment contains numerous large pipe systems networks in 3D space, time
savings in generating realistic installation schedules for pipe systems is of great significance. Moreover, traditional
approach does not consider reliable knowledge data available and cannot guarantee optimal, sequential instillation
schedule. The installation schedule generated using the proposed framework is optimal, sequential, and utilizes
hybrid knowledge data based system to generate efficient and realistic installation schedule.
5. CONCLUSIONS AND FUTURE WORK
A practical installation schedule at any instance of time is a crucial aspect of the piping project in achieving
efficient installation flow, high productivity, and completion of project at a reduced cost and shorter duration.
Many schedules have numerous conflicting features which are only detected during the installation stage.
Therefore, realistic planning of pipe installation schedules with a better understanding of the potential constraints,
sequencing logic, and timely update of required information on pipe system installation is important. In this study
a new AI-based framework with ML and heuristic optimizing techniques is proposed for automating the
optimization of practical installation schedule for pipe systems using 4D BIM. The BIM model is used as a
reference for providing necessary geometric and semantic information of pipe systems and components. KNN and
GNN ML are adopted and compared to integrate extracted components with the appropriate activities and
resources productivity rates for installation based on the developed hybrid knowledge based. GNN performed
better than KNN giving shorter schedule times. The hybrid knowledge base considered both data-driven
knowledge base and site-driven knowledge base. Spatial constraint analysis with local boundary search was
developed to find a reliable clash-free installation sequence for pipe systems. SA algorithm was adopted to
optimize the installation schedule based on spatial constraint analysis and information from the hybrid knowledge-
based system. The framework was tested using a hypothetical piping model and the results show that the proposed
AP Error PR Error
AP Error PR Error
PS8 (3) AP2 0.5 PR2 0.38
PS9 (1) AP5 0.5 PR2 0.38
PS9 (2) AP6 0.2 PR2 0.38
PS9 (3) AP7 0.13 PR2 0.38
PS9 (4) AP6 0.2 PR2 0.38
PS9 (5) AP4 0.13 PR2 0.38
PS10 (1) AP3 0.43 PR2 0
PS10 (2) AP4 0.13 PR2 0
PS10 (3) AP2 0.2 PR2 0
PS11 (1) AP5 0.2 PR1 0
PS11 (2) AP3 0.43 PR2 0
PS11 (3) AP2 0.5 PR3 0
PS12 (1) AP5 0.5 PR1 0
PS12 (2) AP2 0.5 PR2 0
PS12 (3) AP5 0.5 PR2 0.38
PS13 (1) AP2 0.5 PR2 0
PS13 (2) AP5 0.5 PR2 0.38
PS13 (3) AP7 0.13 PR2 0
TABLE 5. Recommended activity packages and productivity rates as per GNN technique
Recommended Activity Packages (AP) Recommended Productivity Rates (PR)
Pipe system (Pipe Number)
AP Error PR Error